Questions to reviewers¶
This is second submission. I got a lot of help from first review, thanks!!!
Still, i have a few questions:
- the first reviewer mentioned Xavier weight initialization; but i didn't find how to use it in code. is it already in tensorflow?
- the first reviewer recommended "low drop_rate", i used 0.5 here. did you mean we should drop little and keep much?
- the first reviewer recommended D: 64 -> 128 -> 256, G:512->256->128->color channel; here i used d: 14x14x64 -> 7x7x128 -> 4x4x256 and g: 7x7x512 -> 14x14x256->28x28x128->28x28x channel. at last step of generator i didn't change image size. did my converions make sense? what are the normally accepted hyper parameters in this project?
- my mentor told me that i might try (strides=1, padding='valid') to convert 4x4 to 7x7; he said it is all about data manipulation; which i agree. i remember in previous videos it was mentioned that in general padding='same' works; so here i stick with 'same'. should i also try 'valid', while focusing on image sizes? // i can try various parameters by myself, but i am not sure how much i should generalize my observations. so here i am asking you. thanks!
- my training results don't look like the best to me, especially for face generation, i occassionally see ghost faces; and it seems to me that the training cannot produce ideal results in near epochs. is it expected for our project? what we may be able to do if training does not go well as expected?
- may we use different learning rate for D and G? here we use drop out for D, we worry about its overfitting; if so, may we use a smaller learning rate for D?
- we use beta1 in AdamOptimizer; we divide all images into multi batches; does this mean we train the model on different batches using different learning rate? // the project code first create a placeholder for learning rate; but later pass a float learning rate into optimizers; so the placeholder is not really used. my question is: is it reasonable to go back to use a larger learning rate at the beginning of each epoch -- with the hope of jumping out of local optima-- and then inside each epoch the learning rate gets smaller and smaller? does anyone do this? ... ... thanks!
Face Generation¶
In this project, you'll use generative adversarial networks to generate new images of faces.
Get the Data¶
You'll be using two datasets in this project:
- MNIST
- CelebA
Since the celebA dataset is complex and you're doing GANs in a project for the first time, we want you to test your neural network on MNIST before CelebA. Running the GANs on MNIST will allow you to see how well your model trains sooner.
If you're using FloydHub, set data_dir to "/input" and use the FloydHub data ID "R5KrjnANiKVhLWAkpXhNBe".
data_dir = './data'
# FloydHub - Use with data ID "R5KrjnANiKVhLWAkpXhNBe"
#data_dir = '/input'
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper
helper.download_extract('mnist', data_dir)
helper.download_extract('celeba', data_dir)
show_n_images = 25
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
import os
from glob import glob
from matplotlib import pyplot
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
print(mnist_images.shape)
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')
print(mnist_images[0].max())
CelebA¶
The CelebFaces Attributes Dataset (CelebA) dataset contains over 200,000 celebrity images with annotations. Since you're going to be generating faces, you won't need the annotations. You can view the first number of examples by changing show_n_images.
show_n_images = 25
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
celeba_images = helper.get_batch(glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB')
pyplot.imshow(helper.images_square_grid(celeba_images, 'RGB'))
print(celeba_images.shape)
print(celeba_images[0].max())
Preprocess the Data¶
Since the project's main focus is on building the GANs, we'll preprocess the data for you. The values of the MNIST and CelebA dataset will be in the range of -0.5 to 0.5 of 28x28 dimensional images. The CelebA images will be cropped to remove parts of the image that don't include a face, then resized down to 28x28.
The MNIST images are black and white images with a single color channel while the CelebA images have 3 color channels (RGB color channel).
Build the Neural Network¶
You'll build the components necessary to build a GANs by implementing the following functions below:
model_inputsdiscriminatorgeneratormodel_lossmodel_opttrain
Check the Version of TensorFlow and Access to GPU¶
This will check to make sure you have the correct version of TensorFlow and access to a GPU
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf
# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer. You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))
# Check for a GPU
if not tf.test.gpu_device_name():
warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))
Input¶
Implement the model_inputs function to create TF Placeholders for the Neural Network. It should create the following placeholders:
- Real input images placeholder with rank 4 using
image_width,image_height, andimage_channels. - Z input placeholder with rank 2 using
z_dim. - Learning rate placeholder with rank 0.
Return the placeholders in the following the tuple (tensor of real input images, tensor of z data)
import problem_unittests as tests
def model_inputs(image_width, image_height, image_channels, z_dim):
"""
Create the model inputs
:param image_width: The input image width
:param image_height: The input image height
:param image_channels: The number of image channels
:param z_dim: The dimension of Z
:return: Tuple of (tensor of real input images, tensor of z data, learning rate)
"""
# TODO: Implement Function
real_input = tf.placeholder(tf.float32, [None, image_height, image_width, image_channels], name='real_input')
z_data = tf.placeholder(tf.float32, [None, z_dim], name='z_data')
lr = tf.placeholder(dtype=tf.float32, name='learning_rate')
return real_input, z_data, lr
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_inputs(model_inputs)
Discriminator¶
Implement discriminator to create a discriminator neural network that discriminates on images. This function should be able to reuse the variables in the neural network. Use tf.variable_scope with a scope name of "discriminator" to allow the variables to be reused. The function should return a tuple of (tensor output of the discriminator, tensor logits of the discriminator).
def discriminator(images, reuse=False):
"""
Create the discriminator network
:param images: Tensor of input image(s)
:param reuse: Boolean if the weights should be reused
:return: Tuple of (tensor output of the discriminator, tensor logits of the discriminator)
"""
# copy the structure from DCGAN code: first layer output has not batch normalization
alpha = 0.2
dropout_rate = 0.5
#print('in discriminator: ', images.shape)
with tf.variable_scope('discriminator', reuse=reuse):
x1 = tf.layers.conv2d(images, 64, 5, strides=2, padding='same')
#bn1 = tf.layers.batch_normalization(x1, training=True)
bn1 = x1
relu1 = tf.maximum(alpha * bn1, bn1)
d1 = tf.layers.dropout(relu1, rate=dropout_rate)
# 14 * 14 * 64
x2 = tf.layers.conv2d(d1, 128, 5, strides=2, padding='same')
bn2 = tf.layers.batch_normalization(x2, training=True)
relu2 = tf.maximum(alpha * bn2, bn2)
d2 = tf.layers.dropout(relu2, rate=dropout_rate)
# 7 * 7 * 128
x3 = tf.layers.conv2d(d2, 256, 5, strides=2, padding='same')
bn3 = tf.layers.batch_normalization(x3, training=True)
relu3 = tf.maximum(alpha * bn3, bn3)
d3 = tf.layers.dropout(relu3, rate=dropout_rate)
# 4 * 4 * 256
flatten = tf.reshape(d3, [-1, 4 * 4 * 256])
logits = tf.layers.dense(flatten, 1)
output = tf.sigmoid(logits)
return output, logits
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_discriminator(discriminator, tf)
Generator¶
Implement generator to generate an image using z. This function should be able to reuse the variables in the neural network. Use tf.variable_scope with a scope name of "generator" to allow the variables to be reused. The function should return the generated 28 x 28 x out_channel_dim images.
def generator(z, out_channel_dim, is_train=True):
"""
Create the generator network
:param z: Input z
:param out_channel_dim: The number of channels in the output image
:param is_train: Boolean if generator is being used for training
:return: The tensor output of the generator
"""
# TODO: Implement Function
alpha = 0.2
#print('in generator ', z, out_channel_dim, is_train)
with tf.variable_scope('generator', reuse= not is_train): # 'not is_train', don't play this trick
x1 = tf.layers.dense(z, 7 * 7 * 512)
x1 = tf.reshape(x1, [-1, 7, 7, 512])
bn1 = tf.layers.batch_normalization(x1, training=is_train)
relu1 = tf.maximum(alpha * bn1, bn1)
# 7 * 7 * 512
#print(relu1.shape)
x2 = tf.layers.conv2d_transpose(relu1, 256, 5, strides=2, padding='same')
bn2 = tf.layers.batch_normalization(x2, training=is_train)
relu2 = tf.maximum(alpha * bn2, bn2)
# 14 * 14 * 256
#print(relu2.shape)
x3 = tf.layers.conv2d_transpose(relu2, 128, 5, strides=2, padding='same')
bn3 = tf.layers.batch_normalization(x3, training=is_train)
relu3 = tf.maximum(alpha * bn3, bn3)
# 28 x 28 x 128
#print(relu3.shape)
logits = tf.layers.conv2d_transpose(relu3, out_channel_dim, 5, strides=1, padding='same')
#print(logits.shape)
output = tf.tanh(logits)
return output
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_generator(generator, tf)
Loss¶
Implement model_loss to build the GANs for training and calculate the loss. The function should return a tuple of (discriminator loss, generator loss). Use the following functions you implemented:
discriminator(images, reuse=False)generator(z, out_channel_dim, is_train=True)
def model_loss(input_real, input_z, out_channel_dim):
"""
Get the loss for the discriminator and generator
:param input_real: Images from the real dataset
:param input_z: Z input
:param out_channel_dim: The number of channels in the output image
:return: A tuple of (discriminator loss, generator loss)
"""
# TODO: Implement Function
#print('model_loss: ', input_real, input_z, out_channel_dim)
g = generator(input_z, out_channel_dim, is_train=True)
d_real, d_real_logits = discriminator(images=input_real, reuse=False)
d_z, d_z_logits = discriminator(images=g, reuse=True)
d_loss_real = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=d_real_logits, labels=tf.ones_like(d_real)*0.9))
d_loss_z = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=d_z_logits, labels=tf.zeros_like(d_z)))
d_loss = d_loss_real + d_loss_z
g_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=d_z_logits, labels=tf.ones_like(d_z)))
return d_loss, g_loss
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_loss(model_loss)
Optimization¶
Implement model_opt to create the optimization operations for the GANs. Use tf.trainable_variables to get all the trainable variables. Filter the variables with names that are in the discriminator and generator scope names. The function should return a tuple of (discriminator training operation, generator training operation).
def model_opt(d_loss, g_loss, learning_rate, beta1):
"""
Get optimization operations
:param d_loss: Discriminator loss Tensor
:param g_loss: Generator loss Tensor
:param learning_rate: Learning Rate Placeholder
:param beta1: The exponential decay rate for the 1st moment in the optimizer
:return: A tuple of (discriminator training operation, generator training operation)
"""
# TODO: Implement Function
vs = tf.trainable_variables()
vs_d = [v for v in vs if v.name.startswith('discriminator')]
vs_g = [v for v in vs if v.name.startswith('generator')]
#serious, i don't know what this is doing; and didn't have time going into it
with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):
d_train_opt = tf.train.AdamOptimizer(learning_rate=learning_rate, beta1=beta1).minimize(d_loss, var_list=vs_d)
g_train_opt = tf.train.AdamOptimizer(learning_rate=learning_rate, beta1=beta1).minimize(g_loss, var_list=vs_g)
return d_train_opt, g_train_opt
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_opt(model_opt, tf)
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np
def show_generator_output(sess, n_images, input_z, out_channel_dim, image_mode):
"""
Show example output for the generator
:param sess: TensorFlow session
:param n_images: Number of Images to display
:param input_z: Input Z Tensor
why pass a z tensor ?
TypeError: The value of a feed cannot be a tf.Tensor object. Acceptable feed values include Python scalars, strings, lists, numpy ndarrays, or TensorHandles.
:param out_channel_dim: The number of channels in the output image
:param image_mode: The mode to use for images ("RGB" or "L")
"""
cmap = None if image_mode == 'RGB' else 'gray'
z_dim = input_z.get_shape().as_list()[-1]
example_z = np.random.uniform(-1, 1, size=[n_images, z_dim])
samples = sess.run(
generator(input_z, out_channel_dim, False),
feed_dict={input_z: example_z})
images_grid = helper.images_square_grid(samples, image_mode)
pyplot.imshow(images_grid, cmap=cmap)
pyplot.show()
Train¶
Implement train to build and train the GANs. Use the following functions you implemented:
model_inputs(image_width, image_height, image_channels, z_dim)model_loss(input_real, input_z, out_channel_dim)model_opt(d_loss, g_loss, learning_rate, beta1)
Use the show_generator_output to show generator output while you train. Running show_generator_output for every batch will drastically increase training time and increase the size of the notebook. It's recommended to print the generator output every 100 batches.
global samples
samples = []
global losses
losses = []
print(losses)
mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))
celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
m100=next(mnist_dataset.get_batches(100))
c100=next(celeba_dataset.get_batches(100))
print(m100.min(), m100.max())
print(c100.min(), c100.max())
m100 = m100 * 2
c100 = c100 * 2
print(m100.min(), m100.max())
print(c100.min(), c100.max())
def train(epoch_count, batch_size, z_dim, learning_rate, beta1, get_batches, data_shape, data_image_mode):
"""
Train the GAN
:param epoch_count: Number of epochs
:param batch_size: Batch Size
:param z_dim: Z dimension
:param learning_rate: Learning Rate
:param beta1: The exponential decay rate for the 1st moment in the optimizer
:param get_batches: Function to get batches
:param data_shape: Shape of the data
:param data_image_mode: The image mode to use for images ("RGB" or "L")
"""
# TODO: Build Model
#saver = tf.train.Saver()
#z0 = np.random.uniform(-1, 1, size=(1, z_dim)) # np.random has no get_shape()
z0 = tf.random_uniform((4, z_dim), -1, 1, dtype=tf.float32)
#samples, losses = [], []
losses=[]
image_channels = 1 if data_image_mode == 'L' else 3
print(' set up in train() ', data_shape[0], data_shape[1], image_channels, z_dim)
print(' in train(), data_shape: ', data_shape)
real_input, z_data, lr = model_inputs(data_shape[1], data_shape[2], image_channels, z_dim)
d_loss, g_loss = model_loss(tf.cast(real_input, tf.float32), z_data, image_channels)
d_train_opt, g_train_opt = model_opt(d_loss, g_loss, learning_rate, beta1)
step = 0
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
for epoch_i in range(epoch_count):
step = 0
for batch_images in get_batches(batch_size):
# TODO: Train Model
#print(' a batch', batch_images.shape)
#print(' data ', batch_images[0]) # so, the real_input is already normalized!
batch_images = batch_images * 2
batch_z = np.random.uniform(-1, 1, size=(batch_size, z_dim))
_ = sess.run([d_train_opt, g_train_opt], feed_dict={real_input:batch_images, z_data:batch_z, lr:learning_rate})
#w2 = sess.run(g_train_opt, feed_dict={real_input:batch_images, z_data:batch_z, lr:learning_rate})
step += 1
if step % 100 == 0:
d_loss_, g_loss_ = sess.run([d_loss, g_loss], feed_dict={real_input:batch_images, z_data:batch_z, lr:learning_rate})
losses += [(d_loss_, g_loss_)]
print(d_loss_, g_loss_, step*batch_size, epoch_i+1)
show_generator_output(sess, 4, z0, data_shape[-1], data_image_mode)
# print('...', step)
#print(w1, w2)
# print eval
show_generator_output(sess, 36, tf.random_uniform((36, z_dim), -1, 1, dtype=tf.float32), data_shape[-1], data_image_mode)
show_generator_output(sess, 36, tf.random_uniform((36, z_dim), -1, 1, dtype=tf.float32), data_shape[-1], data_image_mode)
#saver.save(sess, './checkpoints/generator.ckpt')
return losses
MNIST¶
Test your GANs architecture on MNIST. After 2 epochs, the GANs should be able to generate images that look like handwritten digits. Make sure the loss of the generator is lower than the loss of the discriminator or close to 0.
%%time
batch_size = 32
z_dim = 128
learning_rate = 0.0002
beta1 = 0.2 # 0.1 -- 0.3
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 3
with tf.Graph().as_default():
train(epochs, batch_size, z_dim, learning_rate, beta1, mnist_dataset.get_batches,
mnist_dataset.shape, mnist_dataset.image_mode)
CelebA¶
Run your GANs on CelebA. It will take around 20 minutes on the average GPU to run one epoch. You can run the whole epoch or stop when it starts to generate realistic faces.
%%time
batch_size = 32
z_dim = 128
learning_rate = 0.0001
beta1 = 0.1
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 2
celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
celeba_dataset.shape, celeba_dataset.image_mode)
Submitting This Project¶
When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_face_generation.ipynb" and save it as a HTML file under "File" -> "Download as". Include the "helper.py" and "problem_unittests.py" files in your submission.